【作者】 叶天语；

【导师】 钮心忻；

【作者基本信息】 北京邮电大学 ， 信号与信息处理， 2009， 博士

【摘要】 信息隐藏技术有着广泛的应用,日益受到重视。本文针对信息隐藏理论与算法进行了深入的研究,包括信息隐藏容量、水印检测、强鲁棒性水印算法、自嵌入脆弱水印算法、多功能双水印算法和可逆水印算法几个方向。本文的主要贡献有：(1)针对被动攻击下的信息隐藏系统,本文建立一个具体的信息隐藏统计模型,从安全性、失真距离度量、信道容量三个角度全面分析该模型。信道容量问题的求解抽象成一个受到平均汉明失真距离和平均安全性约束的最优化问题,只要给定需要的参数就可以运用最优化理论得到最优解。最后求解出原始载体序列和秘密信息序列均服从均匀分布时的统计模型信道容量。(2)量化投影(Quantized Projection,QP)技术是量化索引调制(Quantization Index Modulation,QIM)技术工作在投影域的版本,实质上是一个双边加性嵌入器。本文提出符号量化投影(Sign Quantized Projection,SQP)的概念。不同于传统QP,SQP是一个单边嵌入器。理论分析表明：SQP和QP具有相同的漏警概率和虚警概率。同时,SQP的文档水印比(Document-to-Watermark Ratio,DWR)比QP小10log10(1+γ)。(3)打印-扫描攻击会带给数字图像许多干扰,因此抗打印-扫描攻击的数字水印算法对鲁棒性的要求非常高。目前,绝大多数鲁棒水印算法无法抵抗打印-扫描攻击。本文利用离散小波变换(Discrete Wavelet Transformation,DWT)低频逼近子带的能量聚集效应和奇异值分解(Singular Value Decomposition,SVD)的奇异值稳定性,提出一种抗打印-扫描鲁棒零水印算法。原始图像的DWT低频逼近子带进行分块SVD,通过判断两个相邻子块最大奇异值的大小关系产生零水印。实验结果表明算法能够抵抗去同步攻击和打印-扫描攻击。(4)针对一些脆弱水印算法的篡改敏感性需要进一步改进和无法区分图像恶意篡改类型这些不足之处,本文提出一种基于SVD和混沌映射的自嵌入脆弱认证水印算法,将混沌映射的初值敏感性应用到图像篡改检测,并且实现区分图像恶意篡改类型功能。对图像子块进行SVD,同时利用Logistic混沌映射产生认证水印；然后利用Logistic混沌映射设计了一个子块与子块之间的映射函数,将认证水印序列嵌入在相应映射块的最低有效位(Least Significant Bit,LSB)上。实验结果表明：算法不仅对篡改非常敏感,而且能够有效地抵抗拼贴攻击和区分篡改图像内容与篡改水印。(5)针对单水印算法往往存在功能单一的问题,本文利用奇异值的稳定性,提出一种多功能双水印算法。先在图像分块的奇异值上嵌入鲁棒水印,然后在含鲁棒水印图像的空域LSB嵌入脆弱水印,并设计了判别恶意篡改和无意篡改的准则。实验不仅考察鲁棒水印抵抗攻击的鲁棒性,而且还考察脆弱水印对鲁棒性的影响和篡改检测与定位的能力。实验结果表明：鲁棒水印具备很强的抗攻击鲁棒性；脆弱水印对篡改敏感,而且篡改定位精确。因此算法具备版权保护和内容认证双重功能。(6)不同于传统的失真补偿量化索引调制(Distortion-Compensated Quantization Index Modulation,DC_QIM),本文提出了动态DC_QIM的概念。两者的区别在于量化步长和DC参数是否可变。首先推导出动态DC QIM可逆性的成立条件,然后推导出动态DC参数的可允许范围,最后利用可逆性设计一个具体的可逆水印算法。在只执行一遍时,它的数据隐藏率高达1比特每像素,高于其他可逆水印算法。另外,算法的动态特性有利于防止参数泄露。实验结果表明：不管初始条件如何,该算法既能正确解码出秘密信息,又能准确复原原始载体。更多还原

【Abstract】 Information hiding technology has broad applications, and has been paid more and more attention. The theory and algorithm of information hiding are deeply studied in the paper, including information hiding capacity, watermark detection, watermark algorithm with strong robustness, self-embedding fragile watermark algorithm, multi-purpose dual watermark algorithm and reversible watermark algorithm.The attributions of the paper are:(1) With respect to information hiding system under passive attacks, a concrete statistical model of information hiding system is built and its performance is analyzed, including security, distortion distance metric, and channel capacity. The solution to channel capacity can be regarded, subjected to average hamming distortion distance and average security coefficient, as an optimization problem. As long as the required parameters are given, the optimization problem of channel capacity will be easily solved by using optimization theory. Finally, the channel capacity of statistical model is derived on the condition that both original cover sequence and secret information sequence obey to uniform distributions.(2) Quantized projection (QP) technology is the version of quantization index modulation (QIM) working in the projection domain. It essentially belongs to a double-sided additive embedder. The concept of sign quantized projection (SQP) is proposed in the paper, which differs from the conventional QP in working as a single-sided embedder. Theoretical analysis reveals that SQP has the same probability of miss and probability of false alarm as QP. In the meanwhile, the document-to-watermark ratio (DWR) of SQP is less than its counterpart of QP by 10log10(1+γ)dB.(3) The print-and-scan process always brings a lot disturbance to the original digital image, therefore the watermarking algorithm for print-and-scan process desires extremely high robustness. At present, most robust watermarking algorithms can’t resist the print-and-scan process. A robust zero-watermark algorithm for print-and-scan process is proposed in the paper, based on the energy aggregation effect of discrete wavelet transformation’s (DWT) low frequency band and the stability of singular value decomposition’s (SVD) singular values. It conducts DWT with the original image, divides its low frequency band into blocks, and conducts SVD with every block. Zero-watermark is derived by judging the numerical relationship between two biggest singular values from adjacent blocks. The experimental results show that it has the ability to resist desynchronization attacks and print-and-scan process.(4) At present, the tamper sensitivity of some fragile watermark algorithms needs to be improved. Moreover, some fragile watermark algorithms can’t distinguish image’s malicious tamper type. A self-embedding fragile authentication watermark algorithm based on SVD and chaotic mapping is proposed, trying to apply chaotic mapping’s initial value sensitivity to tamper detection and distinguish image’s malicious tamper type. SVD is carried out on each image block, and authentication watermark is produced by using Logistic chaotic mapping. Then Logistic chaotic mapping is used to design a mapping function defined on two separate blocks, and authentication watermark sequence is embedded into the corresponding mapping block’s least significant bit (LSB). The experimental results show that the algorithm is not only very sensitive to tamper, but also resists collage attack and distinguishes image content tamper from watermark tamper.(5) The single watermark algorithms always merely have single function. In order to overcome the drawback, a multi-purpose dual watermark algorithm is proposed in the paper, taking advantage of the stability of singular values. It divides the original image into several blocks, and inserts robust watermark into their singular values. Then it inserts the fragile watermark into LSB in the space domain of previous watermarked image. Moreover, it designs a rule to distinguish malicious tamper from unintentional tamper. It tests not only robust watermark’s robustness towards attacks, but also fragile watermark’s influence towards robustness as well as the ability to tamper detection and localization. The experimental results show that the robust watermark possesses strong robustness to resist attacks, and that the fragile watermark is very sensitive to tamper and has good accuracy of tamper localization. Therefore it can achieve copyright protection and content authentication at the same time.(6) The concept of dynamic distortion-compensated quantization index modulation (Dynamic DC-QIM) is proposed, which differs from the conventional DC-QIM in whether or not quantization step size and DC parameter are changeable. Firstly, the condition satisfying the reversibility of dynamic DC-QIM is deduced. Then, the allowable range of dynamic DC parameter is derived. Finally, the procedure of reversible watermark algorithm based on dynamic DC-QIM is designed. Its data hiding rate can achieve as high as lbpp in a single iteration, higher than its former counterparts. Furthermore, the use of dynamic characteristic is in favor of preventing its parameters from being disclosed. Experimental results show that it not only accurately decodes the secret information, but also perfectly restores the original cover, regardless of initial conditions.更多还原